High Throughput Connectomics
summary
Project aims to produce petabyte-to-gigabyte reductions by extracting connectomes at the pace of imaging for a cubic millimeter (two petabytes) and ultimately layout graphs rather than full imagery. Pipeline uses an XNN execution framework on a 72‑core CPU and targets roughly four hours per terabyte (about 2× microscope speed); membrane/vesicle/cleft detectors plus a synapse rule find multi-contact strong synapses. New algorithms include mask‑extend, flood‑fill comparisons and a 3C cross‑classification clustering that parallelizes seed extensions; results run fast but still show split/merge errors needing richer context.
AI-written summary · about